Location Privacy without Carrier Cooperation
Keen Sung
Brian Neil Levine
Marc Liberatore
School of Computer Science, Univ. of Massachusetts Amherst
{ksung,brian,liberato}@cs.umass.edu
Abstract—Cellular network operators can track the location
of cell phone users as they connect to different towers.
Operators may not directly control the user’s phone, but they
do supply and control the SIM card that identifies the user.
We seek to preserve a cellular phone user’s location privacy
from cellular network operators. We propose the ZipPhone
protocol for secure, virtual, and therefore easily changeable
SIM cards. ZipPhone breaks the association between the user
and IMSI identifier, and thus prevents the cellular operator
from localizing the user. At the same time, it still allows
authentication, billing, and E911 service by the operator. We
empirically analyze the effectiveness of ZipPhone against a
passive carrier. This class of attacker has a location profile of
the user before they switched to ZipPhone, but relies on the
normal operation of GSM mechanisms to learn the location of
users. We reproduce the results of a previous inference study
and show that it did not realistically model GSM carriers.
We show that ZipPhone users can expect to be deanonymized
only 6% of the time, which is a sixth of the rate reported by
previous work.
I. I NTRODUCTION
Mobile device users depend on the cellular phone and
data network infrastructure on a continuous basis. Cellular
network operators can track cell phone users as they travel
among and connect to towers, violating their location privacy.
In this work, we propose a scheme to preserve users’
location privacy by allowing them to easily change virtual
SIM cards over the Internet. This scheme, which we call
ZipPhone1 , is backwards-compatible with existing cellular
infrastructure.
The difficulty of preserving the location privacy of
cellphone users is that an entire infrastructure of densely
placed radio towers can observe the phone’s signals. And
although the cellular carrier doesn’t necessarily control the
phone hardware, they do supply and control the SIM card
carried by the user in their phone. SIM cards contain a small
computer, protected by tamper-resistant packaging, able
to process cryptographic functions and carrying a unique
identifier. Despite this scenario, we believe location privacy
is still possible.
We detail a method for obtaining location privacy without
the active cooperation of the carriers that control the
cellular infrastructure. Rather than obfuscating the location
of the user, our approach is to break the one-to-one
link between user and SIM card, as the latter is what
carriers use to identify users of localized handsets. As we
will demonstrate, ZipPhone can provide location privacy
from current carriers because their defenses are aimed at
mitigating cloning, which is the unauthorized duplication
1 Our system is named for its parallels to the Zipcar rental system. In
ZipPhone, the user doesn’t own the SIM, she rents it only temporarily as
she moves around town.
of the credentials stored in the SIM card; ZipPhone lends
credentials rather than cloning them. If carriers targeted
their defenses on ZipPhone the problem would be harder,
since they might deny service to phones that appear to be
running ZipPhone, or they might localize phones beyond
normal GSM operation.
At a high level, ZipPhone works as follows. ZipPhones
do not have a SIM card. Instead, they access a remote
set of credentials to instantiate a virtual SIM locally. The
credentials are accessed using Wi-Fi or other networks not
observable to the carrier. Once a virtual SIM is obtained,
the usual GSM protocols are still able to handle switching
among towers, refreshing virtual SIMs, and other details.
In sum, our contributions are as follows.
• We propose a backwards-compatible method of GSM
location privacy that does not rely on the cooperation of
the carrier. New credentials can be re-issued relatively
frequently over the air. Our primary design relies on
a cooperative but untrusted mobile virtual network
operator (MVNO) that issues ephemeral identities and
session keys. We also propose a variant of this system,
which uses peer-based exchange of SIM credentials,
and relies on flaws in the GSM system first discovered
in 2001 and never fixed. We call this variant PeerPhone.
• We propose a method of thwarting a class of active
attacks on users that reveal their location. Such attacks
are generally based on GSM pages for fake phone calls
or SMS messages. Our solution is an application of
portknocking [34].
• We empirically analyze the effectiveness of ZipPhone
against a passive carrier. This class of attacker has a
location profile of the user before they switched to
ZipPhone, but relies on the normal operation of GSM
mechanisms to learn the location of users. We reproduce the “conclusive” results of a previous inference
study [41] and show that it does not realistically model
GSM carriers. We show that ZipPhone users can expect
to be denonymized only 6% of the time, which is less
than a sixth of the rate reported by previous work.
• We discuss other attacks against ZipPhone. First, we
detail how carriers can determine that a phone is using
ZipPhone and therefore deny it service. Second, we
discuss active attacks based on localizing the phone
without their cooperation (e.g., lateration attacks).
In the next two sections, we review the operation of GSM
networks and define our security model. We then present the
details of ZipPhone, discuss its validity in terms of carrier
terms-of-service, analyze its security, and place it in the
context of related work.
2
II.
S UMMARY OF GSM N ETWORKS
Cellular carriers use one of a few technologies to support
mobile phones. In this section, we give an overview of
the details of the GSM system and terminology [54], and
the aspects of the GSM architecture that are required for
our discussion of ZipPhone. ITM-2000 based systems have
differences but largely follow the same general architecture;
GSM is the foundation of GPRS, WCDMA, EDGE, UMTS,
and LTE. Moreover, most networks support reverting to a
GSM mode.
A phone handset is known in GSM as a mobile station
(MS), and each is composed of its mobile equipment (ME)
and its subscriber identity module (SIM) card. To connect,
the handset connects via its radio interface to a radio
tower, which is called a base transceiver station (BTS).
A single base station controller (BSC) can manage many
BTSs, including functions related to resource and mobility
management. The BTS and BSC form the base station
subsystem (BSS).
A. Connecting to the Network
Each BSC is connected to one mobile switching center/visitor location register (MSC/VLR); see Fig. 1. The
MSC/VLR controls call setup and routing, among other
tasks. One MSC/VLR connects to many BSCs. Each
MSC/VLR is also connected to the carrier network’s
home location register (HLR), which records the particular
MSC/VLR where each phone subscriber may be found. The
HLR is associated with an authentication center (AuC) that
stores cryptographic credentials needed for communicating
with each of the carrier’s SIM cards.
Each SIM card contains an international mobile subscriber identity (IMSI) and a unique symmetric key Ki . The
same IMSI and key pairing is stored in the AuC. Generally,
SIMs do not allow querying of the key. Each ME contains
an international mobile equipment identity (IMEI) number,
that is comprised of a unique serial number and the make,
model, and place of manufacture of the phone.
When a phone is powered on, it scans for BTS towers
available from the carrier specified in the SIM, each on a
different frequency. To connect to the network, the phone
begins a location update process by selecting the tower with
the strongest signal and then requesting a communications
channel from the BTS. As part of the process, the handset
sends its IMSI to the MSC via the BTS. (If requested, the
handset will also send its IMEI.) The MSC will request
authentication information from the HLR/AuC with a mobile
application part (MAP) send authentication info message for
this particular IMSI. MAP is the internal signaling protocol
for GSM nodes.
To generate a response, the HLR/AuC selects a random
number and combines it with the IMSI’s Ki to generate
a session key (Kc ) and a signed response (SRES), using
a given algorithm. The three values (called a triplet) are
returned to the MSC/VLR. The MSC forwards, via the BTC,
only the random number to the handset. The handset asks
its SIM to generate the signed response that corresponds to
the random number (for the given algorithm), and forwards
the value to the MSC/VLR via the BTS. The SIM also
generates Kc for the phone from the random number.
AuC
HLR
MSC/
VLR
MSC/
VLR
BSC
BSC
BTS
Fig. 1: The simplified GSM architecture.
If the value matches, the handset is authenticated, and
the MSC sends a MAP update location message to the
HLR. The HLR associates the IMSI with the address of
the MSC/VLR. The MSC then tells the BTS the session
key, and then the phone and BTS switch to an encrypted
channel. The MSC assigns and sends the phone a temporary
mobile subscriber identity (TMSI) that is unique to the
location area it resides in. Although the TMSI is sent
encrypted, it can be retrieved from the SIM on some phones
and there are scenarios where it is broadcast over the
network unencrypted [35]. Finally, the phone releases the
communications channel. It is now camped on a cell and is
ready to use the network’s services. For example, incoming
calls from the public switched telephone network are routed
through the HLR, then the MSC/VLR, and after a broadcast
page, from the BTS to the handset.
B. Moving to a new location area
As the user moves within the cells that comprise a location
area, it can use the same TMSI to access the network. Once
the handset moves to a new location area, it must begin
the location updating process again, though it can send its
TMSI (and old location area) instead of the IMSI. If the
TMSI value is already known to the MSC/VLR, the location
update will be accepted without re-authentication. A new
TMSI will be assigned, unique to the new location area.
If the location area was managed by a different MSC/VLR,
then the handset’s credentials must be reacquired. It is
preferred for the MSC to request the associated IMSI
from the previous MSC/VLR, rather than directly from
the handset. The new MSC/VLR also retrieves a triplet
from the HLR and may ask the phone to re-authenticate
with a new signed response.
C. Data access
GPRS provides data service for mobile phones, and it
is the basis for UMTS architectures [3]. The protocol for
activating data service is called a GPRS attach. The process
is similar to a location update, and we only sketch the
details. After requesting and receiving a channel, the MS
sends an attach request to its BSS. The BSS forwards the
request to a serving GPRS support node (SGSN), which
fetches authentication triplets from the HLR. Once the
3
MS authenticates itself based on knowledge of Ki , an
encrypted channel is started between the SGSN and the
mobile. The SGSN exchanges information with the HLR
about the authenticated handset, and the attach is accepted
and acknowledged to the handset, including issuance of a
packet temporary subscriber identity (P-TMSI). The process
will be repeated if the handset moves to a location area
controlled by a different SGSN. UMTS data access is
essentially the same in terms of the signaling to the handset
and the authentication mechanisms.
III. S ECURITY M ODEL
Our security model consists of users seeking location
privacy from cellular carriers. Users carry and control
phones that are equipped with cellular and Wi-Fi radios, and
thus have IP-based data connections on both radios (once
connected). They are allowed to use encryption, which we
assume is reasonably strong. Users are not trying to gain
unauthorized service.
Our model includes carriers that are passive or active
attackers.
Passive Attacker. A passive carrier controls the network
infrastructure, accounts, and SIM cards. They can ensure that
only authorized accounts are used to connect to the network.
They know the incoming and outgoing calls of each phone
and can observe data packets sent via the phone’s cellular
connection (but not the contents of encrypted traffic). The
passive attacker has records of which towers a user’s phone
has associated with during voice and data transfers, and
knows the location of each of its towers and its coverage
area. The attacker cannot force the phone to answer requests
(e.g., pages). The carrier and SIM cards make use of GSM
protocols including the A5 family of ciphers and their
existing weaknesses. We also allow the passive attacker to
leverage past information about user geographic movements
to build a profile.
Networks with an always-update policy expect phones to
perform a location update whenever the phone enters a new
cell; operators never enforce such a policy because it has the
highest overhead for the carrier [48], [61]. The standard is
for phones to perform a forming location area (LA) update
[48], [61], where a phone initiates a location update only
if its location area changes. In that case, the carrier will
learn the exact cell that the phone is associated with only
when it (i) performs the location update or GPRS attach,
(ii) connects to a BTS to make a call or send data, and
(iii) answers pages for incoming calls or data. The passive
attacker does not falsely or unnecessarily prod the user into
performing a location update or similar attacks.
Active Attacker. The active carrier has all the abilities of
the passive attacker and can attempt real-time geolocation
a specific phone, including unnecessarily asking the phone
to initiate a location update process and multilateration of
the received phone signal.
We do not consider certain other active attacks due to
either their fragility or their extreme cost when applied
on a network-wide scale. For example, we do not allow
either attacker to use cameras or other visual information,
nor to physically stalk the user. Because the carrier does
not control the phone hardware or OS in our model, only
the SIM, it cannot, for example, insert SSL keys onto the
phone or otherwise change the phone’s software. Similarly,
the carrier cannot prevent Wi-Fi connections, nor capture
information from the phone’s Wi-Fi, key presses, camera,
GPS, or screen.
Handset signatures. Phone hardware has been shown
to carry many unique signatures that can be determined
remotely, including the radio’s power amplifiers [45], [46]
and the phone’s accelerometer [15]. Although these attacks
can be effective, they are not the focus of this paper. To
obtain privacy, users need both the protocol we propose
presently, and defenses against remote inference of physical
signatures.
IV. Z IP P HONE
Cellular users are subject to several vulnerabilities that
reveal their geographic position, though some are easily
mitigated. First, phones store a unique IMEI, which is akin
to a MAC address; however, this identifier can be modified
by the user since she controls the handset hardware2 . Second,
each user is likely identifiable by the unique set of outgoing
calls they make; however, they can make calls via VOIP
rather than using the cellular carrier. Encryption of the VOIP
stream can thwart carrier eavesdropping. Stronger protection
is available by using VOIP over Tor3 . Third, as our past
work has shown, their location may been inferred from
throughput characteristics, [55], though such attacks are
limited, currently.
A naive privacy solution is offered by some mobile
virtual network operators (MVNOs), such as TracPhone,
Straight Talk, and Boost Mobile. MVNOs have no cell
tower infrastructure. Instead, they buy service wholesale
from carriers and resell to customers. Many MVNOs allow
users to purchase a so-called “burner phone” and a SIM
with pre-paid minutes, which can be disposed of by the
user; such purchases and activations do not require a name
or address. This approach to privacy works well for those
willing to sacrifice a great deal of convenience and money
for privacy, but it does not scale to improving the location
privacy of every day cellular users that are seeking not to
be “observed in all matters” [51].
In the remainder of this section, we identify vulnerabilities
that stem from the use of a specific IMSI (stored on a
tamper-proof SIM) and the geographic consistency of IMSIs
(which in practice are typically tied to a home, workplace,
or both). We then detail our approach to mitigating these
vulnerabilities.
A. Authentication and Location Updating
The most naive approach to obfuscating the relationship
between a user and a geographic trace of a phone by a
cellular network is purchasing a pre-paid SIM card from an
MVNO such as TracPhone, and replacing both it and the
phone after a short period of time. A slightly better approach
is for users to meet up in a cafe and randomly switch SIM
cards and phones. Of course, neither approach is remotely
2 See https://www.blackphone.ch/ for a recent attempt at a handset
designed around consumer privacy.
3 See http://torfone.org/
4
HLR
AuC
HLR
Gateway
AuC
MSC/
VLR
BSC
BSC
MVNO
M
Internet/
TOR
p2p SIM
exch.
MNO/Carrier
phones/users
BTS
Fig. 2: The ZipPhone architecture for authentication and
attachment to the network.
convenient. Further, such a scheme is not secure unless we
can quantify how often a user should swap hardware.
ZipPhone allows for convenient exchange of logical
cellular identifiers. It is based on phones without local
SIM cards; instead, the phone’s software retrieves a virtual
SIM offered by Internet-accessible third party; see Fig. 2.
Carriers that control infrastructure, such as Verizon
and AT&T, are generally called mobile network operators
(MNOs). They market and sell service directly to consumers.
MNOs also have a large wholesale business servicing
MVNOs, which are explicitly allowed to resell service from
an MNO. An MVNO has no national BSS infrastructure,
but may issue SIMs and operate an HLR and AuC. About
100 MVNOs operate in the US alone [60].
ZipPhone is based on a centralized MVNO solution,
though the operator is not trusted with identities. Below, we
also detail a variant called PeerPhone which is decentralized.
ZipPhone is a more secure solution than PeerPhone, but
the former has the disadvantage that a business relationship
with a major carrier is required.
Authentication. Any existing or new MVNO that operates
an HLR and AuC could easily support the ZipPhone architecture. We assume that such an organization is cooperative (it
does not launch denial-of-service attacks) but is not trusted
by the ZipPhone users with their locations or identities.
Let’s call one such MVNO Zipline. Like any MNVO
today, Zipline customers roam on the network of a partnering
MNO. There are two key differences between Zipline and
current MVNOs on the market.
First, Zipline customers are not issued physical SIM
cards; instead they anonymously request ephemeral IMSIs
and IMEIs from Zipline’s SIM Exchange during bootstrap
(e.g., via Wi-Fi). IMSIs are ideally issued as nonces, but
could be reused over time. IMSI requests are paid for (over
Tor) via a suitably anonymous currency such as Bitcoin [42]
or Zerocoin [39] in which no consistent ID is needed. Using
Tor increases the difficulty of a SIM holder inferring that
two ZipPhone requests originated from the same user. We
can assume the leases are for some number of minutes at
a time. The MVNO will know when the lease is initiated
because it will receive an authentication request from the
MNO. To terminate the lease, Zipline can send a message
to the MNO’s VLR, as per current GSM standards that exist
to support the existing pre-paid plans of MVNOs.
Second, Zipline customers prefetch the correct answer to
the next several SRES values. SRES prefetching has two key
advantages: (i) it removes any long delays during signaling
that identifies the phone as participating in ZipPhone,
preventing expiration of GSM timer T3260 [2] and reducing
the possibility of detection by the carrier; (ii) it ensures
that the phone can perform location updates while in the
field without a separate Internet connection. New purchases
and prefetched triplets can be made anonymously over the
data channel even if out of range of the initial Wi-Fi base
station.
No aspect of Zipline requires any changes to the GSM
architecture. While it does require cooperation of a sponsoring MNO, what we propose does not violate the terms of
wholesale service of major carriers to their MVNOs. It is not
necessary for the MNO to know that the MVNO is running
ZipPhone. It is not out of the ordinary for an MVNO to
run its own HLR/AuC. If a current MVNO implemented
this architecture, a partnering MNO would be unable to
easily distinguish customers with physical SIMs from those
leasing remote SIMs.
Location Updates. Conveniently, as the user moves to new
BTS towers and even new location areas, the TMSI will be
refreshed without re-authenticating, as noted in Section II-B.
The carrier will require the phone to re-authenticate only
when a location area operated by a new MSC or SGSN is
reached. These regions can be large, as cells range from
about 1 to 100 km2 , one BTS manages several cells, and
one MSC (or SGSN) manages several location areas. When
re-authentication is required, ZipPhone users use prefetched
values.
To make and receive phone calls, we assume the ZipPhone
user has registered with an (anonymous) VOIP service,
as described in Section IV-C. A significant limitation of
ZipPhone is that the user cannot utilize the legacy phone
system and reasonably expect to retain location privacy.
However, E911 service, which is tied to a handset and not
a user or SIM, would be still available if needed.
B. PeerPhone: Peer-based exchange of SIM credentials
To our knowledge, no current US-based MNO or MVNO
allows the resale of their services by their end-users4 .
Customers of these operators may still wish to obfuscate
their movement by occasionally lending their SIM to others
for use without recompense. We call this variant scheme
PeerPhone, and differentiate it from ZipPhone as needed in
the remainder of this paper.
Lending an authorized phone and SIM, virtually or
physically, does not appear to be against the Terms of
Service of major carriers. This lending makes sense only
for persons with unlimited data and call plans.
Authentication. A user that has been lent a SIM must be
able to answer the challenge-and-response authentication
4 Verizon’s consumer Terms of Service: http://www.verizonwireless.
com/b2c/support/customer-agreement. AT&T’s: http://www.att.com/
shop/en/legalterms.html?toskey=wirelessCustomerAgreement;
TMobile’s:
http://www.t-mobile.com/templates/popup.aspx?PAsset=
Ftr Ftr TermsAndConditions&print=true.
5
protocol posed by the carrier: PeerPhone relays the challenge
to the SIM owner over Wi-Fi. This approach is successful
because the GSM SIM security protocols require and
ensure only that the SIM is authorized to use the network
(as described in Section II-A), rather than requiring it is
physically connected to the handset. The remote SIM holder
is able to provide session key Kc and the SRES responses
to the phone. Because Ki does not need to be passed, the
SIM’s use can subsequently be lent out to another user.
Relaying the carrier’s packets across the Internet via
Tor to the SIM holder will introduce significant delays.
However, because carriers expect to work with resourcepoor phones and do not wish customers to find it hard to
connect to the network, the GSM protocol is not stringent.
The GSM specifications set a 12-second timeout by default
on receiving a response to authentication requests within
a location update process (GSM timer T3260 [2]). If the
timeout occurs, the user can simply try again.
The termination of the remote use of a SIM (and its
IMSI) is easily controlled by the SIM owner, who need
only connect to the carrier and perform a location update.
In doing so, the HLR of the carrier will remove access
from the previous or current MSC/VLR. A SIM owner can
leave an extra Internet-connected phone at their house or
other preferred location (perhaps with the actual SIM) and
instruct it to connect to the network remotely; this setup
allows the owner to reclaim without necessarily revealing
their current location.
Retrieving Credentials from a SIM. Users that wish to
participate in PeerPhone and lend out use of their SIM must
retrieve the Ki key stored in it, though it only needs to be
done once. They can then produce the Kc and SRES values
for a remote PeerPhone requester, who can relay via Wi-Fi
(or existing cellular connection) the random number issued
by the carrier to the peer during a location update. Keys for
encryption with GPRS/EDGE are also based on knowledge
of Ki and can be similarly relayed. The peer should never
relay the value Ki .
To recover Ki , peers can launch one of many known
plaintext- and ciphertext-only attacks [5], [10], [47] against
the various A5 encryption algorithms used by carriers.
Barkan et al. [5] show how to recover the key in seconds
from small amounts of ciphertext for both voice and GPRS
communication for A5/1, A5/2, and A5/3. Almost all
networks use A5/1 or no encryption at all (which is called
A5/0) [52], [56]. (A5/2 is so weak that it is no longer allowed
on GSM networks.)
All communications and all algorithms on a GSM phone
are based on Ki ; as long as carriers allow any portion of
their communications using A5/1, the peer can recover the
key by bidding down from the stronger A5/3 algorithm
to A5/1 during negotiation (which is the basis of many
commercial surveillance products [4]). In December 2013,
Deutsche Telekom upgraded from A5/1 to A5/3, but still
allows breakable connections via A5/1 from phones [40],
and to date it has announced no changes to its US-based
operation, T-Mobile. AT&T has announced it will upgrade
to “parts” of its network, but because it will still allow A5/1
connections, peers can recover their Ki values [56].
By handing out Kc , the remote PeerPhone user can use
the same attacks to recover Ki , but that vulnerability is
already exploitable by all attackers that are within radio
range of the SIM holder; it is no weaker to lend Kc to
remote parties than it is to use the carrier’s network in the
first place. A5/1 was first shown to be weak in 2001 [10] —
more than a decade ago — but this revelation has resulted in
no deployment changes by carriers. Notably, smart phones
on the market are close to having more resources than the
PCs first used to carry out these attacks in 2001 (128 MB
RAM and 146 GB storage [10]).
Location Updates. A PeerPhone can use its TMSI values to
move among BTSs and location areas. When a location area
is reached that is controlled by a different MSC or SGSN,
the phone must reauthenticate. In that case, the PeerPhone
must connect over the existing cellular data connection to
the SIM holder. In some cases, the PeerPhone will be able
to use its existing connection on the old tower to relay
packets back to the SIM holder. To do so, it will need to
switch its frequency several times to act as a relay; most
phones can do so quickly as they are constantly switching
frequencies to test received signal strength (to determine
if a handoff is necessary, even during a voice call). If this
back-and-forth cannot be completed reliably in practice, a
PeerPhone can be engineered to consist of better antennas
to allow more time for the relayed location update process
to take place. In the worst case, a phone can be built with
two radios to avoid the delay of frequency switching. In
this preliminary work, we have not evaluated the efficacy
of these mechanisms.
C. Preserving Privacy During Communication
By initiating and receiving overt GSM or unencrypted
VOIP calls, ZipPhone users risk being identified via a
profile of call records held by the carrier. Some protection
is gained from using an encrypted VOIP service since it
would not reveal to the carrier the identity of the user, whom
she calls, or from whom she receives calls. If the VOIP
service itself cannot be trusted, then an anonymous VOIP
service is required, such as Torfone. Anonymous VOIP has
a performance penalty [36].
An additional threat is posed by several attacks that
request the user’s phone to silently associate with a real
or unauthenticated tower [22] based on an SMS Class 0
message, ICMP ping, or similar technique. The general
approach of these attacks is to page the phone, and once
the phone is associated to a cell, the call is abandoned. It’s
easy for ZipPhones to ignore all incoming calls and SMS
messages, but since legitimate incoming VOIP calls are first
received as pages for incoming data packets, a similar attack
exists based on pages for GPRS data.
Thwarting Fake Cellular Pages. We propose an application of portknocking [14], [34] to defend against these
attacks. The defense, which we call pageknocking has the
limitation that it is easily subject to a denial of service
attack by the carrier, and the user can’t deny that it is using
the defense (and therefore also using ZipPhone).
We assume that the user is registered to an Internetbased proxy service that accepts incoming VOIP calls and
anonymously forwards the calls to the user. (We assume the
registration process also creates the necessary entry in the
carrier NAT table.) To signal to the user that an incoming
6
packet is genuine, the service uses a notification coded in
a series of intervals between pages. More generally, this
defense can be used for all incoming GPRS traffic to the
phone if a proxy, VPN concentrator, or Tor node understands
the protocol. In fact, a limitation of the approach is that,
because unrelated incoming traffic can corrupt the signal,
all traffic to the peer should be routed through one Internetbased proxy (or Tor-based service).
The phone waits for incoming VOIP calls via pages on
the GPRS tunneling protocol (GTP). The phone waits in
standby mode, which allows the carrier to know its location
area for routing data, but not the specific cell nearest to the
phone. Ahead of time, the service and the user agree upon
a key and a parameter n. The key can be renegotiated at
any time but we refer to the current key as g. A sequence
of n bits is chosen as the first n bits of the hmac [6] of
the key and the current time t, synchronized to the current
minute: hmac(g, t). The bit string determines a sequence
of transmits and pauses. For each bit that is set, the service
transmits a data packet, which results in a GTP page to
the phone, and then the service pauses for a duration of d
seconds. For each bit that is clear, a packet is not set and
the service pauses for d seconds. The duration of d must be
long enough to ensure that a new page will be generated by
the carrier when a new packet comes in, but not so long that
the chances of another packet being received is significant.
The chances that the carrier (or non-malicious third party)
can generate a false n-bit page is 2−n . At the receiver end,
when a page is received at time t0 , the user calculates the
first n bits of hmac(g, t0 ), and holds t0 constant until the
pattern doesn’t match. If they do match, the phone responds
to the page.
The value of d limits the number of incoming VOIP
calls (and data packets generally) that a user can receive
per minute. However, since the pattern changes once per
minute, the number of chances that a third party has to
brute force the value g is extremely limited. The protocol
does not limit the amount of communication that the user
can initiate.
al. assume an always-update location management policy in
which the user performs a location update with the carrier
for every cell it enters.
As we show, a more realistic model of carrier location
management based forming LA updates (see Section III)
predicts such identification attacks are much less successful
than previously reported. For example, when users change
identifiers once an hour, attackers succeed only 6% of the
time.
The code for our simulations is available from http://
traces.cs.umass.edu.
V. PASSIVE ATTACK A NALYSIS
In this section, we characterize the effectiveness of the
passive attacker against ZipPhone (and PeerPhone). We
ask, What percentage of the time can the passive attackers
determine the identity of a ZipPhone user, given a profile
of that user trained using overt data?
Past work is pessimistic about the success of methods for
location privacy that are based on changing identifiers. For
example, Mulder et al. [41] previously examined the accuracy of techniques that infer the identity of a cellular user
given a month-long training period and varying frequency
of changing IDs. In that paper, they expected that IDs could
be changed as infrequently as once a month or as often as
once an hour. Attacker identification rates were as high as
88% for monthly changes and 48% for hourly changes; the
paper remarks that “this work conclusively demonstrates
that removing identifiers from location information, or
merely blurring the spacial resolution, does not eliminate
the danger of deanonymization.” Below, we reproduce their
classification method and evaluated the same data set [17],
and then perform a critical additional experiment. Mulder et
Self-transitions are ignored (i.e, P r(kp,p ) = 0). We include
a smoothing factor by adding 1 to each of the counts of all
possible transitions before computing Eq. 1.
During testing, a window of M minutes of data from a
single user is randomly selected from the held-out test data
set. A contiguous sequence of cells θ = x1 , x2 , . . . , xn ; xi ∈
S, consisting of n cells visited by an unknown user, is
taken in chronological order from within that window. The
sequence of n − 1 transitions is evaluated against the values
stored in the matrix for each user. An indicator is defined
for each user k as the product of all probabilities in the
associated training matrix. The user with the maximum
value of Ik is selected by the classifier.
A. Empirical Analysis of Passive Attackers
Given the ubiquity of cell phones, it is reasonable to
expect the carrier to already have a profile of its users.
Therefore, our evaluation of ZipPhone in the presence
of a passive attacker assumes that the training data is
easily accessible. We cross-validated our results using three
test/train month pairs, encompassing the majority of the
Eagle et al. [17] data set. The data includes the location
areas, cells, and a record of calls made by 106 users between
November 2004 and February 2005.
Because ZipPhone allows for users to obtain new IMSI
values over the air, it’s reasonable to expect that they
can change identifiers fairly frequently. We have not
implemented ZipPhone, and so we cannot guarantee any
particular frequency. But for the sake of comparison, we
show results for a frequency as (impractically) low as once
a minute.
Inference Algorithm [41]. The inference approach designed by Mulder et al. is summarized as follows. During
training, a square transition matrix Pk is created for each
user k. For a set of all states S visited by all users, the
matrix encodes the probabilities that a user in cell p moves
to cell q for all p, q ∈ S.
P r(kp,q ) = P
Count(p → q)
0
q 0 ∈S\p Count(p → q )
arg max Ik =
k
n−1
Y
P r(kxi ,xi+1 )
(1)
(2)
i=1
While this model does not consider stationary location
data (i.e. self-transitions), Mulder et al. found this approach
more successful than a more sophisticated Markovian model
that we did not reproduce.
7
80%
●
●
Attack Accuracy
70%
●
●
60%
Always−update
Forming LA
Location Area Only
●
●
50%
40%
●
●
30%
●
●
20%
●
●
10%
●
●
0%
●
●
●
1 month
1 week
1 day
1 hr
1 min
Trace Duration (minutes)
Fig. 3: The accuracy of the attack defined by Mulder et al. [41]
under the always-update policy (top, red line) and forming LA
policy (middle, green line). Our results match well with [41]:
an attacker achieves a 38% success rate against users that
update their SIM-based identifiers once an hour. Under the
latter, more realistic, forming location area update policy, the
attacker’s success rate falls to 6% when SIM-based identifiers
are updated once an hour. The bottom, blue line shows the
lower bound on any scheme: it represents an unrealistic
location management scheme where the carrier learns only
the location area but not the cell a user is associated with.
Errorbars represent 95% c.i.
The results for a forming update location management
policy are shown as the middle, green line in Figure 3.
For a one-month sequence of locations — corresponding
to changing IDs via ZipPhone once a month — accuracy
falls by more than half, as compared to always-updating,
to about 34%. For a one-hour sequence of cell transitions,
accuracy is at 6%. As an impractical lower bound, we also
show results for assigning new IMSI IDs after each cell
transition. In that case, the attacker will succeed with less
than 1% accuracy.
The bottom, blue line shows the lower bound for this
inference technique at each trace duration: it represents an
unrealistic location management scheme where the carrier
learns only the location area but not the cell a user is
associated with.
Limitations. We focus on the algorithm in Mulder et al.
because it is a simple approach that is designed for GSM
networks, and a universally accepted result. In future work,
we plan to evaluate additional inference algorithms. The
Eagle et al. data is relatively small; it has the advantage
of containing cell IDs and call records for individual users
(c.f., taxi cab data with no user-level consistency or call
records). But we seek to expand our evaluation of ZipPhone
to larger data sets. Finally, we hope to implement a version
of ZipPhone using OpenBSC to understand the costs and
limitations of frequent identifier updates. Additional attacks
on ZipPhone are discussed in the following section.
VI. D ISCUSSION : OTHER ATTACKS
ZipPhone is vulnerable to at least two other major sets of
attacks6 , which we discuss in this section: enumeration of
all phones participating in ZipPhone; and active localization
by carriers against phones in-between forming LA updates.
In this preliminary work, we discuss each of these attacks,
and leave empirical analysis for future work.
Results. The effectiveness of the classifier is shown in
Figure 3. In all cases, the attacker is given a preceding
month’s data as ground truth for training. The top, red line
is a recreation of results from Mulder et al.: a randomly
selected 1-month-long sequence of the cells a user is
associated with results in a high accuracy of 80%; Mulder
et al. saw5 about 82%. A random sample of up to 1-hour
of cell locations is identifiable 38% of the time; Mulder et
al. saw about 44%. In both cases, random chance would be
correct about 1% of the time.
Mulder et al.’s results are pessimistic in that they are
for an always-update scenario, which is never implemented
by carriers in practice [48], [61]. The attacker’s accuracy
is much lower for a more realistic forming update policy,
where the carrier only learns the exact cell that the phone
has associated with when the mobile device performs a
location update — the device performs an update only upon
entering a new location area or receiving a call (rather than
upon entering each cell within a location area), or when the
carrier initiates a connection. Fortunately, Eagle et al.’s data
includes a record of calls for each phone. In our simulations,
we conservatively assume that phone call duration is an
average of 3 minutes long [62].
A. Identifying SIMs Participating in ZipPhone
In many but not all cases, a carrier can determine which
SIMs on its network are running ZipPhone or the PeerPhone
variant. Once identified, the carrier may elect to deny service
to these users. In sum, the centralized protocol (e.g., an
MVNO such as Zipline, running the ZipPhone protocol) is
immune to most simple attacks, yet does not require the
user trust Zipline.
First, the carrier can attempt to enumerate all SIM cards
that are participating in the system. Enumeration is possible
in PeerPhone if all venues where peers advertise their
interest in lending credentials can be found and joined.
This enumeration is not possible with an MVNO, unless it
repeats IMSI values or colludes with a carrier.
Second, the carrier can compare the latency of a handset
in responding to network signaling to others on the network.
It should be significantly greater if a remote SIM is actually
being contacted to complete authentication signaling. In the
ZipPhone MVNO-based case, the values can be prefetched,
defeating this attack; in PeerPhone exchange, they cannot
be prefetched.
5 These small differences are due to our use of an additional month from
the data set.
6 Another risk we do not discuss is the physical danger of assuming
another user’s credentials [50].
8
Third, the carrier can observe the locations (i.e., cells
and sectors) where a SIM card is being used and look for
SIMs that follow a geographic pattern that is not typical, for
example, jumping from city to city or country to country
faster than a person can reasonably travel. PeerPhone might
make this attack easier because the SIM will always return
to its home location when a lending session terminates —
unless the user only uses the SIMs of others via a fair
exchange protocol. Zipline would not be subject to this
attack since it has the option of never reusing an IMSI.
Fourth, a carrier can isolate those users that never receive
a phone call and always use VOIP. In the MVNO case,
again this analysis might not be effective if the IMSI is
never used again, but would still raise suspicion depending
on the length of time the IMSI is active in the network.
As a defense, a user could place phone calls occasionally
to arbitrary numbers (e.g., libraries or businesses) but we
don’t consider such a defense here.
Finally, whether in the peer-based or MVNO scenario,
any phone that is engaging in the pageknocking defense
against an active attacker described in Section IV-C will be
fairly obvious.
Only AT&T and T-Mobile have deployed U-TDoA. (Most
European operators use Cell ID.)
In practice, there are limits on the carrier’s ability to
track all users via multilateration. Ficek [21] has shown
some of these limits. For example, to locate a user via
a false SMS message, a series of signals and the GSM
broadcast paging channel are needed. These resources limit
the number of users that can be tracked per minute to very
low numbers depending on various parameters, e.g., tracking
more than 20 users out of 410 exceeds network capacity.
During the tracking, phone calls and data traffic that rely
on the broadcast channel must be delayed.
In future work, we plan to evaluate the effectiveness of
these two strategies and the cost to the carrier in terms of
broadcast capacity.
B. Active Attacks
Location Based Services. First, papers on privacypreserving location-based services (LBS) generally assume
that the user is submitting queries for service. For example,
the user may query for the nearest restaurant or gas station.
Cellular networks are an LBS in that the user’s query is a
request for mobile voice or data service (rather than content)
from a tower within radio range.
Not all solutions and analyses of privacy-preserving LBS
are easily applied to the cellular scenario. Cellular users
cannot introduce a trusted intermediary to obfuscate the
mobile’s position or introduce fake queries to the carrier [38].
Solutions that assume the user can control the level of
granularity of their location are also not applicable [63].
Past work on deanonymization of private traces of mobile
users assume the user’s pseudonym is unchanged throughout
the trace. A small amount of external information, such as
the person’s home or work address [28], can deanonymize
an obfuscated trace [7], [8], [24], [32], [37], [41] given
a consistent identifier. In contrast, we strive to change
pseudonyms as often as network resources can support,
which may be in minutes. Indeed, work by Zang and
Bolot [65] shows that suitably anonymizing a trace of 25
million cellular users across 50 states (30 billion records
total) requires only that users have the same pseudonym
for no longer than a day. A day’s duration is unsuitable
for Zang and Bolot’s goal of supporting researchers that
wish to characterize the behavior of users over time (while
maintaining their privacy). On the other hand, the result is
promising for users seeking privacy, who might be able to
change their pseudonyms much more frequently than once
per day.
Others solutions can be adapted to our cellular scenario.
A cellular user can use mix zones [8], [23], abstain from
service [9], and introduce some types of false information
[31], [53].
The active attacker’s success depends on the method and
quality of localization used, and the frequency with which
they localize each user.
Most studies of cellular localization assume the phone
itself participates in localization, which is more accurate
(e.g., [13], [58]) but is not available against a ZipPhone user,
according to our attacker model. Other studies assume that
the phone is coerced into revealing its location [16], [21],
[22], but we expect that such an attack can be thwarted by
pageknocking, though we do not evaluate the defense here.
Accordingly, active carriers have two main options:
Localization based on Cell ID. The carrier may localize
the user based on the Cell ID it is associated with. A phone
always knows its Cell ID, but the carrier learns the Cell
ID of a phone only when it uses voice or data services.
Cells themselves can vary greatly in size, from 100 km2 to
1 km2 (or smaller if dedicated to a single home or business).
Trevisani and Vitaletti [57] performed in-field experiments
in the US and found that Cell IDs are offset from the
true location of the mobile by about 0.8 km in an urban
environment (Manhattan), 0.5 km in a suburban environment,
and 2.9 km in a highway environment. Watzdorf and
Michahelles [59] report similar results. Ficek [22] reports
accuracy of 0.05–20 km.
Localization based on multilateration. The carrier may
use the Uplink Time Difference of Arrival (U-TDoA)
method of multilateration. In U-TDoA, the carrier tracks
the time it takes for the same signal to reach its network
of base stations. The error is much lower than Cell ID
at 0.11 km in tests [1] (Ficek [22] reports 0.04–0.12km).
However, deploying U-TDoA requires a massive upgrade
to the network infrastructure [22]. Accordingly, CDMAbased carriers, Verizon and Sprint, have instead deployed
assisted GPS (aGPS), which requires participation from the
phone, and thus must use Cell IDs for passive localization.
VII. R ELATED W ORK
There are hundreds of papers on location privacy [25],
[33] for mobile users, spanning a number of paradigms
from indoor mobile ad hoc networks to outdoor cellular
networks, and from queries for information to friend-finding
services and social network check-ins. Very broadly, the
papers most related to our work fall within two key topics.
Location Privacy for Mobile Users. Secondly, many works
address cellular location privacy. In contrast to our work,
9
most assume the carrier is willing to deploy changes. Some
focus on enlisting a (trusted) carrier to protect against a
third party [20], [26], [27], [35]. Reed et al. [49] propose
privacy from the carrier using onion routing, but does not
consider the direct connection that must be made to a tower.
Federrath et al. [19] propose a similar scheme that prevents
linkability of calls between two parties but omit critical
details regarding authentication to the carrier. Fatemi et
al. [18] propose an anonymous scheme for UMTS using
identity-based encryption, but unlike our approach, that
scheme involves the carrier in the cryptographic exchange;
they enumerate the vulnerabilities of similar works [29],
[44], [64], [67]. The closest work to ours is Kesdogan et
al. [30], which proposes using a trusted third party to create
pseudonyms for GSM users, but also routes all calls through
that provider, which allows it to characterize the calling
pattern and infer the caller.
Anonymous VOIP. The TORFone project (http://torfone.
org/) has implemented a TCP-based voice-over-Tor system.
(Others have examined the feasibility of real-time traffic over
TCP [11], [12], [43], [66].) Our prior work has characterized
the performance of a multi-hop/Tor-like UDP-based voiceover-IP, and shown it performs with sufficient quality of
service despite the three-hop paths [36].
VIII.
R EFERENCES
[1]
3GPP. TS 45.811: Feasibility Study on Uplink TDOA in GSM and
GPRS. http://www.3gpp.org/DynaReport/45811.htm, July 2007.
[2]
3GPP TS 04.08 version 7.21.0. Mobile radio interface layer 3
specification. http://www.etsi.org/deliver/etsi ts/100900 100999/
100940/07.21.00 60/ts 100940v072100p.pdf Default time values on
page 574.
[3]
3GPP TS 23.060 version 11.7.0. Universal Mobile Telecommunications System (UMTS); General Packet Radio Service (GPRS);
Service description. http://www.etsi.org/deliver/etsi ts/123000
123099/123060/11.07.00 60/ts 123060v110700p.pdf.
[4]
Advanced Surveillance Technology, Inc.
Wide-area (spiderweb) passive interception system. media.wix.com/ugd/70bfc5
619d9e6fc11df018f9af893328e798dd.pdf.
[5]
E. Barkan, E. Biham, and N. Keller. Instant Ciphertext-Only
Cryptanalysis of GSM Encrypted Communication. Journal of
Cryptology, 21(3):392–429, Mar. 2008.
[6]
M. Bellare, R. Canetti, and H. Krawczyk. Keying Hash Functions for
Message Authentication. In Proc. Advances in Cryptology, volume
1109 of LNCS, pages 1–15. 1996.
[7]
A. Beresford and F. Stajano. Location privacy in pervasive
computing. IEEE Pervasive Computing, 2(1):46–55, 2003.
[8]
A. Beresford and F. Stajano. Mix zones: user privacy in locationaware services. In Proc. Pervasive Computing and Communications
Wrkshps, pages 127–131, 2004.
[9]
L. Bindschaedler, M. Jadliwala, I. Bilogrevic, I. Aad, J.-P. Hubaux,
V. Niemi, and P. Ginzboorg. Track Me If You Can: On the
Effectiveness of Context-based Identifier Changes in Deployed
Mobile Networks. In Proc. ISOC NDSS, Feb. 2012.
[10]
A. Biryukov, A. Shamir, and D. Wagner. Real Time Cryptanalysis
of A5/1 on a PC. In Proc. Intl. Wrkshp on Fast Software Encryption,
pages 1–18, 2001.
[11]
E. Brosh, S. Baset, V. Misra, D. Rubenstein, and H. Schulzrinne.
The Delay-Friendliness of TCP for Real-Time Traffic. IEEE/ACM
Trans. on Networking, 18(5):1478–1491, 2010.
[12]
E. Brosh, S. A. Baset, D. Rubenstein, and H. Schulzrinne. The
delay-friendliness of tcp. In Proc. ACM SIGMETRICS, pages 49–60,
2008.
[13]
M. Y. Chen, T. Sohn, D. Chmelev, D. Haehnel, J. Hightower,
J. Hughes, A. LaMarca, F. Potter, I. Smith, and A. Varshavsky.
Practical Metropolitan-Scale Positioning for GSM Phones. In Proc.
UbiComp, volume 4206 of LNCS, pages 225–242, 2006.
[14]
R. deGraaf, J. Aycock, and M. J. Jacobson. Improved Port Knocking
with Strong Authentication. Proc. Annual Computer Security
Applications Conf., pages 451–462, 2005.
[15]
S. Dey, N. Roy, W. Xu, R. R. Choudhury, and S. Nelakuditi.
AccelPrint: Imperfections of Accelerometers Make Smartphones
Trackable. In Proc. ISOC NDSS, Feb. 2014.
[16]
K. Dufková, M. Ficek, L. Kencl, J. Novak, J. Kouba, I. Gregor,
and J. Danihelka. Active GSM Cell-id Tracking: “Where Did
You Disappear?”. In Proc. ACM Intl. Wrkshp on Mobile Entity
Localization and Tracking in GPS-less Environments, pages 7–12,
2008.
[17]
N. Eagle and A. Pentland. Reality Mining: Sensing Complex Social
Systems. Personal and Ubiquitous Computing, 10(4):255–268, 2006.
[18]
M. Fatemi, S. Salimi, and A. Salahi. Anonymous roaming in
universal mobile telecommunication system mobile networks. IET
Information Security Journal, 4(2):93–103, 2010.
[19]
H. Federrath, A. Jerichow, D. Kesdogan, and A. Pfitzmann. Security
in Public Mobile Communication Networks. In Proc. IFIP/TC6
Personal Wireless Communications, pages 105–116, April 1995.
[20]
H. Federrath, A. Jerichow, and A. Pfitzmann. MIXes in Mobile
Communication Systems: Location Management with Privacy. In
Proc. Intl. Wrkshp on Information Hiding, pages 121–135, 1996.
[21]
M. Ficek. Tracking Users in Mobile Networks: Data Acquisition
Methods and their Limits. PhD thesis, Czech Technical University
in Prague, https://dspace.cvut.cz/bitstream/handle/10467/18881/
C ONCLUSION
We proposed ZipPhone, a method for obtaining location
privacy without the active cooperation of the carriers that
control the cellular infrastructure. ZipPhone is backwardscompatible and designed to allow new IMSI identifiers to
be re-issued relatively frequently over the air. A cooperative
but untrusted MVNO can easily support ZipPhone by
issuing ephemeral identities and session keys, and it is
less vulnerable to many attacks that are possible against
PeerPhone, a peer-based version of ZipPhone. PeerPhone
can be deployed without the logistics of managing a
relationship with an MNO, but is more vulnerable to attacks.
We also propose a method of thwarting attacks that prod
the user to associate with cells based on fake pages. Our
solution is an application of portknocking.
Our empirical analysis of the ZipPhone’s protection
against passive attacks by a carrier points out the limitations
of results in a previous study [41]. That work assumed
an always-update policy which is too costly for carriers
to implement in practice. Our simulation of a forming LA
update shows that ZipPhone users that update identifiers
hourly can expect to be deanonymized only 6% of the time,
which is less than a sixth of the rate reported by previous
work.
In future work, we plan to quantify ZipPhone’s protection
against other inference algorithms, using additional data
sets. We also plan to quantify the success of attacks that
determine which phones are using ZipPhone (or page
knocking defenses). And we plan to quantify the success of
active attackers that localize the user without the user’s
participation, and quantify the costs to the carrier for
launching such attacks.
Acknowledgements. This work was supported in part by
NSF award CNS-0905349.
10
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
TEZE Diserta%C4%8Dn%C3%AD%20pr%C3%A1ce Ficek
Michal 2013.pdf, June 2013.
M. Ficek, T. Pop, and L. Kencl. Active tracking in mobile networks:
An in-depth view. Computer Networks, 57(9):1936 – 1954, 2013.
J. Freudiger, R. Shokri, and J.-P. Hubaux. On the Optimal Placement
of Mix Zones. In Proc. PETS, pages 216–234, Aug. 2009.
P. Golle and K. Partridge. On the anonymity of home/work location
pairs. In Proc. Intl. Conf. on Pervasive Computing, pages 390–397,
2009.
A. Görlach, A. Heinemann, and W. Terpstra. Survey on Location
Privacy in Pervasive Computing. In Privacy, Security and Trust
within the Context of Pervasive Computing, volume 780 of The Intl.
Series in Engineering and Computer Science, pages 23–34. 2005.
M. Gorlatova, R. Aiello, and S. Mangold. Managing base station
location privacy. In Proc. MILCOM, pages 1201–1206, Nov. 2011.
M. Gorlatova, R. Aiello, and S. Mangold. Managing location privacy
in cellular networks with femtocell deployments. In Proc. WiOpt
Symposium, pages 418–422, May 2011.
S. Isaacman, R. Becker, R. Cáceres, S. Kobourov, M. Martonosi,
J. Rowland, and A. Varshavsky. Identifying Important Places in
People’s Lives from Cellular Network Data. In Proc. Intl. Conf. on
Pervasive Computing, pages 133–151, 2011.
Y. Jiang, C. Lin, X. Shen, and M. Shi. Mutual Authentication and
Key Exchange Protocols for Roaming Services in Wireless Mobile
Networks. IEEE Trans. on Wireless Communications, 5(9):2569–
2577, 2006.
D. Kesdogan, H. Federrath, A. Jerichow, and A. Pfitzmann. Location
Management Strategies Increasing Privacy in Mobile Communication. In Information Systems Security, pages 39–48. 1996.
H. Kido, Y. Yanagisawa, and T. Satoh. An anonymous communication technique using dummies for location-based services. In Proc.
Intl. Conf. on Pervasive Services, pages 88–97, 2005.
J. Krumm. Inference Attacks on Location Tracks. In Proc. Intl.
Conf. on Pervasive Computing, pages 127–143, May 2007.
J. Krumm. A survey of computational location privacy. Personal
Ubiquitous Comput., 13(6):391–399, Aug. 2009.
M. Krzywinski. Port Knocking: Network Authentication Across
Closed Ports. In SysAdmin Magazine, volume 12, pages 12–17.
2003.
D. F. Kune, J. Koelndorfer, N. Hopper, and Y. Kim. Location leaks
on the GSM Air Interface. In Proc. ISOC NDSS, Feb. 2012.
M. Liberatore, B. Gurung, B. N. Levine, and M. Wright. Empirical
Tests of Anonymous Voice Over IP. Elsevier Journal of Network
and Computer Applications, 34(1):341–350, January 2011.
C. Y. Ma, D. K. Yau, N. K. Yip, and N. S. Rao. Privacy vulnerability
of published anonymous mobility traces. In Proc. MobiCom, pages
185–196, 2010.
J. Meyerowitz and R. Roy Choudhury. Hiding Stars with Fireworks:
Location Privacy Through Camouflage. In Proc. Annual Intl. Conf.
on Mobile Computing and Networking, pages 345–356, 2009.
I. Miers, C. Garman, M. Green, and A. D. Rubin. Zerocoin:
Anonymous Distributed E-Cash from Bitcoin. In Proc. IEEE
Symposium on Security and Privacy, pages 397–411, 2013.
Mobile Europe. Deutsche telekom upgrades mobile network security
with a5/3 encryption standard. http://www.mobileeurope.co.uk/PressWire/deutsche-telekom-upgrades-mobile-network-security-witha5-3-encryption-standard, December 2013.
Y. D. Mulder, G. Danezis, L. Batina, and B. Preneel. Identification
via Location-profiling in GSM Networks. In Proc. ACM Wrkshp
on Privacy in the Electronic Society, pages 23–32, 2008.
S. Nakamoto. Bitcoin: A Peer-to-Peer Electronic Cash System.
http://www.bitcoin.org/bitcoin.pdf, 2009.
M. F. Nowlan, N. Tiwari, J. Iyengar, S. O. Aminy, and B. Fordy.
Fitting square pegs through round pipes: Unordered delivery wirecompatible with tcp and tls. In Proc. USENIX NSDI, pages 28–28,
2012.
J. Park, J. Go, and K. Kim. Wireless authentication protocol
preserving user anonymity. In Proc. SCIS, pages 159–164, 2001.
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
A. Polak, S. Dolathshahi, and D. Goeckel. Identifying Wireless
Users via Transmitter Imperfections. IEEE JSAC: Special Issue on
Advances in Digital Forensics for Communications and Networking,
29(7):1469–1479, August 2011.
A. Polak and D. Goeckel. RF Fingerprinting of Users who Actively
Mask Their Identities with Artificial Distortion. In Proc. Asilomar
Conf. on Signals, Systems, and Computers, May 2011.
J. Rao, P. Rohatgi, H. Scherzer, and S. Tinguely. Partitioning attacks:
or how to rapidly clone some GSM cards. In IEEE Symp. Security
and Privacy, pages 31–41, 2002.
S. M. Razavi. Tracking Area Planning in Cellular Networks
[Elektronisk resurs] : Optimization and Performance Evaluation.
Linköping, 2011.
M. G. Reed, P. F. Syverson, and D. M. Goldschlag. Protocols using
anonymous connections: Mobile applications. In Security Protocols,
volume 1361 of LNCS, pages 13–23. 1998.
J. Scahill and G. Greenwald. The NSA’s Secret Role in the U.S.
Assassination Program. https://firstlook.org/theintercept/article/2014/
02/10/the-nsas-secret-role/, 10 Feb 2014.
B. Schneier. The eternal value of privacy. Wired, May 2006.
Security Research Labs. Gsm map project: gsm security country
report (usa). http://gsmmap.org/assets/pdfs/gsmmap.org-country
report-United States of America-2013-08.pdf, August 2013.
R. Shokri, G. Theodorakopoulos, G. Danezis, J.-P. Hubaux, and
J.-Y. Boudec. Quantifying Location Privacy: The Case of Sporadic
Location Exposure. In Proc. PETS, pages 57–76, Aug. 2011.
C. Smith and D. Collins. 3G Wireless Networks. 2nd edition, 2007.
H. Soroush, K. Sung, E. Learned-Miller, B. N. Levine, and
M. Liberatore. Disabling GPS is Not Enough: Cellular location
leaks over the Internet. In Proc. PETS, pages 103–122, July 2013.
C. Timberg and A. Soltani.
By cracking cellphone code,
NSA has capacity for decoding private conversations.
http://www.washingtonpost.com/business/technology/bycracking-cellphone-code-nsa-has-capacity-for-decodingprivate-conversations/2013/12/13/e119b598-612f-11e3-bf4561f69f54fc5f story.html, December 13 2013.
E. Trevisani and A. Vitaletti. Cell-ID Location Technique, Limits
and Benefits: An Experimental Study. In Proc. IEEE Wrkshp on
Mobile Computing Systems and Applications, pages 51–60, 2004.
A. Varshavsky, M. Chen, E. de Lara, J. Froehlich, D. Haehnel,
J. Hightower, A. LaMarca, F. Potter, T. Sohn, K. Tang, and I. Smith.
Are GSM Phones THE Solution for Localization? In Proc. Wrkshp
Mobile Computing Systems and Applications, pages 34–42, 2006.
S. von Watzdorf and F. Michahelles. Accuracy of Positioning Data
on Smartphones. In Proc. Intl. Wrkshp on Location and the Web,
pages 2:1–2:4, 2010.
Wikipedia. List of us mobile mvnos. Last checked 201312-31. http://en.wikipedia.org/wiki/List of United States mobile
virtual network operators.
V. W.-S. Wong and V. C. Leung. Location Management for Nextgeneration Personal Communications Networks. IEEE Network,
14(5):18–24, Sept. 2000.
J. Wortham. Cellphones Now Used More for Data Than for
Calls. http://www.nytimes.com/2010/05/14/technology/personaltech/
14talk.html, May 13 2010.
T. Xu and Y. Cai. Feeling-based Location Privacy Protection for
Location-based Services. In Proc. ACM CCS, pages 348–357, 2009.
G. Yang, D. Wong, and X. Deng. Efficient anonymous roaming
and its security analysis. In Applied Cryptography and Network
Security, volume 3531 of LNCS, pages 334–349. 2005.
H. Zang and J. Bolot. Anonymization of Location Data Does Not
Work: A Large-scale Measurement Study. In Proc. ACM MobiCom,
pages 145–156, 2011.
X. Zhang and H. Schulzrinne. Voice over tcp and udp. Technical
Report CUCS-033-04, Columbia University, 2004.
J. Zhu and J. Ma. A new authentication scheme with anonymity
for wireless environments. IEEE Trans. on Consumer Electronics,
50(1):231–235, 2004.